I’m sitting here waiting for a fluid sim to bake. I’m nearing the 24-hour mark and it’s not even half done. My CPU is an AMD 8350 overclocked to 5Ghz, not that this matters much since the fluid sim rarely pushes my CPU over the 20% point.
Needless to say, this painful process has me thinking about other CPUs, specifically the incredibly powerful cores in Intel’s newest lineup. But while we have tons of rendering benchmarks posted, we don’t seem to have any baking benchmarks.
Is this true? I looked around the archives and couldn’t find anyone creating a smoke/fluid benchmark. Because if Intel chips significantly outperform AMD, I would be very tempted to rip out my whole motherboard and put an Intel kit in.
Since no one responded, I’m assuming that my guess was correct: there is no benchmark for this. I would like to make one. Is anyone else interested in the data? And if so, do you have any input on how to create the benchmark? I’m thinking about the peculiarities of the way Blender handles fluid and smoke that I may not know about, and how to best integrate these considerations into a benchmark.
I’d be interested to see something like this.
I have an Intel Q9400 I run at 3.21 GHz and I’m almost finished building it’s replacement with Intel’s latest and greatest I7-4790k so that should give some interesting comparisons. Just come up with a fluid sim and a smoke sim and post them and I’ll send my results.
If somebody comes up with better sample files it can be changed. Make something you can bake in a couple hours so it doesn’t take forever to test
You’d need some way of effectively measuring the time it takes, I don’t believe the simulations report time spent baking. if you can get over the hurdles, it would be a cool test.
It makes me wonder about doing a full test suite, baking physics, baking textures, gpu/cpu rendering, boolean operations. some sort of scripted operation that measures the overall system performance for blender.
hmm, didn’t realize it doesn’t give you a time stamp, this could be handled with a script like Sterling suggested.
I would mention that I just did a little test and my dual core laptop and quad core desktop both pegged all cores at 100%
You may be using blender without openMP enabled, that makes a huge difference
I was using the latest build from the buildbot
I am using a non-OpenMP build (because all of the 2.70+ builds from Graphicall are unstable on my system).
That’s the biggest issue that I currently see. OpenMP is not part of the official release, but it could make my AMD match an Intel chip with faster, but fewer, cores. Do I include that in considerations even though it’s unofficial. I would lean toward no. But that would remove some of the real-world usefulness from the benchmark.
The problem of timing I think has been solved. All one has to do is look in their cache folder and look at the “date modified” for the first file of the fluid bake and the last file. That gives you the time it took. I also found a python request that sends back a time. I’m going to experiment with that… as soon as my fluid bake is done.
If you’re gonna try doin’ benchmarks for sim-baking, keep one thing in mind:
While your typical rendering-benchmark is usually about rendering a still (not an animation), baking a physics-sim means Blender will write a file to disc after simulating each frame.
The point is: In your usual rendering-benchmark no data is written to disc, and even if, then just once in the end. Thus the time it takes to write the framebuffer to disc is neglectible.
Now if you’re baking a sim, it’ll write the data to disc afer each frame, so the time it takes your hdd to write this data adds to the raw computing-time your cpu has to spend.
In short, if you want to do sim-baking benchmarks, you should probably consider checking what hdd you’re using, maybe gather some technical data about it (how fast does it write according to the manufacturer?).
I’m by no means trying to imply you could kind of calculate the time it takes to write the data to disk and then substract this from your benchmark’s results in order to know the raw computing-time. This would quite sure not work out, be confusing and rather useless anyway.
But I think it might make a notable difference if s.o. were e.g. using the exact same system for your benchmark, only once writing the files to an hdd at 5400 upm and once one at 7200 upm.
In any case I heared it makes a big difference weather you bake to an hdd or an ssd (the latter is obviously much faster).
So to sum it all up: Other than knowing what CPU you used 4 baking, it’d be probably good to know about the hdd either, to get usefull, comparable results from this.
I hadn’t thought of that, but I think that I can effectively “smudge” this variable out of consideration.
As long as the benchmark is measured in multiple minutes, I think its usefulness is maintained. Even on a slow hard drive, the fluid sim files would take hundredths of a second at most to write to the hard drive. That means that a 100-frame bake would only add a few seconds. If the benchmark is designed to take at least five minutes, the consequences of writing to the hard drive fall into the margin of error.